[JVM-B] Re: Thanks for your classwriter modules, and small bug? in getfield/putfield

[Ashley Y. <as...@se...>]

At 2002-10-11 08:13, Arthur Baars wrote:

>Hi Ashley,
>
>I found your JVM-bridge software after reading the "Java representation in
>Haskell" thread on the haskell mailing list. I used your java class file
>writer(JCF*.hs) as a back-end for a simple "Featherweight Java" compiler.
>It really like the library, it was easy to use to assemble class files.
>
>Unfortunately the class file writer modules did not work using Hugs. The
>main reason is that most of them import "BasicLayer", which in itself
>imports other non-hugs compatible modules. Fortunately none of the JCF
>modules really needs the BasicLayer(some of them require Bytes.hs, but
>that's all). After changing the imports I only had a problem with the Word64
>type which is not fully implemented in Hugs. Removing all the code that
>referred to the Word64 type, and adapting the "JFCWritable" instance for
>"Int64" was sufficient to get all your class writer code running in Hugs.

I haven't even tried to use it with Hugs...

>I added some code as well to the instruction module, because I needed the
>new, checkcast and dup instructions.

OK, I've put these in (in CVS).

>Are you planning to add all jvm instructions to JCFCodeInstruction in the
>future?

Originally I added just enough to write callback classes. I can't add 
branch instructions without rewriting the stack depth calculation code, 
etc.

>Maybe it is a good idea to distribute the class file writer as a
>separate library as well.

This is an excellent idea. I'd like to do a new release shortly, and 
after that I want to organise the modules hierarchically. All the stuff 
that doesn't use FFI (the 'Basic', 'Java' and 'JCF' layers) except maybe 
JCFCallback* should be in a separate project that can be made to work 
with Hugs.

>I think I discovered a small bug in your code in the getfield and putfield
>instructions.

Thanks, I've put this fix in.

-- 
Ashley Yakeley, Seattle WA